Phone sequence modeling with recurrent neural networks

In this paper, we investigate phone sequence modeling with recurrent neural networks in the context of speech recognition. We introduce a hybrid architecture that combines a phonetic model with an arbitrary frame-level acoustic model and we propose efficient algorithms for training, decoding and sequence alignment. We evaluate the ad- vantage of our phonetic model on the TIMIT and Switchboard-mini datasets in complementarity to a powerful context-dependent deep neural network (DNN) acoustic classifier and a higher-level 3-gram language model. Consistent improvements of 2–10% in phone accuracy and 3% in word error rate suggest that our approach can readily replace HMMs in current state-of-the-art systems.